We Need to Pump the Brakes on Driverless Cars
We’re nowhere near proving that automated driving technology is safer than human drivers
Back in late 2016, Michigan became one of the first states to pass laws explicitly governing the testing and deployment of automated vehicles (AVs). Part of that package of new rules included the ability to test vehicles on public roads without a safety driver. Numerous states now have similar regulations in place, but they all have one thing in common. Nowhere in the United States are AV companies required to prove to local officials that their vehicles are actually safe to operate on public roads.
Training and evaluating humans who want to get a driver’s license in the United States is severely lacking compared to many other countries, especially those in Europe. Often the most challenging test that a teenager has to perform to get a permit to operate a two-ton vehicle is parallel parking. Demonstrating that they can adequately respond to an obstruction in the road and safely stop or steer clear is not part of the evaluation. That said, people do have to take a basic vision test and demonstrate that they understand the rules of the road.
In the United States, no such requirement exists for software-powered virtual drivers. While tens of millions of miles have been racked up by development AVs over the past decade along with billions of simulation miles, we are nowhere near close to demonstrating that automated driving technology is even as safe as human drivers, much less safer.
For some reason, the United States has decided that it’s okay to submit the public to a large-scale experiment that no one consented to or was even informed of.
Prior to the economic slowdown triggered by the pandemic, Americans were driving more than 3.2 trillion miles annually. While 38,000 fatalities a year on our roads is absolutely a tragedy, given how much we drive, that works out to about 1.1 deaths every 100 million miles. While human error plays a part in upwards of 90% of crashes, we still only crash about every half a million miles. That works out to about once every 30 years at an average of 15,000 miles per year.
Why Dead Sonos Speakers Mean You’ll Never Own a Driverless Car
Tesla has changed the way we see cars and software
The only state that publicly reports any notable data about AV testing is California, with its well-publicized but deeply flawed system disengagement reports. Even the companies with the lowest rates of disengagement (when humans are required to take control of an AV) are still doing it less than every 20,000 miles.
For some reason, the United States has decided that it’s okay to submit the public to a large-scale experiment that no one consented to or was even informed of. Those hundreds of AVs being tested in Silicon Valley, Phoenix, Miami, Pittsburgh, and Ann Arbor don’t even have to meet the absurdly low standards that humans do.
I bring all this up because I recently had the opportunity to go for a ride around downtown Ann Arbor in a prototype AV. Over the past dozen years I’ve ridden in many AVs from different companies in a variety of locations, starting in 2008 with the Chevrolet Tahoe from Carnegie Mellon University that won the 2007 DARPA Urban Challenge. Until January 2020, every one of those vehicles that was on the road had a human safety operator. The only truly driverless experiences I’d had were on closed test tracks.
As much as I trust and respect the people at many of these companies, I remain unconvinced that they should be allowed to operate without some independent supervision.
Over the years I’ve come to know and respect many of the people at these companies. I’ve looked at their processes and I’m confident that they are taking due care with training the vehicle operators and putting processes in place to minimize the risk inherent in the work they do.
This has not always been the case. Uber, prior to March 2018, was responsible for a particularly egregious case that led to the death of Elaine Herzberg. That incident was a combination of both human and technical failures. As much as I trust and respect the people at many of these companies, I remain unconvinced that they should be allowed to operate without some independent supervision.
In China, AV companies must pass a series of objective evaluations before they’re granted permits to test on public roads. A multi-phase graduated system defines where they’re allowed to test based on the vehicle’s capabilities. When changes are made to the vehicle configuration, such as replacing a sensor with a different type or moving the sensor to a new location, these tests must be redone before the revised vehicles can go on the road.
Swarms of Teeny Robo-Tractors Will Outmaneuver Tesla’s Driverless Cars
While Elon Musk and Waymo get all the attention, autonomous vehicles for farming face fewer tech barriers and could be…
Before Mobileye (owned by Intel) could begin testing in Germany recently, it had to go through an independent evaluation by TÜV Süd, one of the technical monitoring organizations that handles the vehicle type approval process there. In the United States, AV companies file some paperwork with the state and pay a fee and then go about their testing.
During my most recent Ann Arbor ride, there was no safety operator behind the wheel. A representative of the company sat next to me in the back seat (both of us masked) with a remote disable switch sitting next to him. The ride was mostly, but not entirely, uneventful. However, at one four-way stop, the car slowed and rolled through without coming to a complete halt.
At another intersection where the car approached as the traffic signal transitioned from yellow to red, it could not seem to make a decision and ended up stopping in the middle of the intersection as the signal went red. The company rep had to hop out and get into the driver’s seat to take control. Fortunately, University of Michigan students had not yet returned to the city, so traffic was lighter than usual.
All of this happened at low enough speeds. Even if a crash had occurred, it probably would not have been life-threatening. However, the reality is that this technology is not yet ready for prime time. Allowing the companies that hope to profit from autonomous vehicles to make the final decision about safety readiness is like allowing a teenager to decide when they are ready to drive solo.
Since the National Highway Traffic Safety Administration under the current administration is unlikely to impose any new regulations, states that allow AV testing should be developing a plan to evaluate the effectiveness of this technology before there is another Elaine Herzberg.